esafak

13 ms·

361.
by esafak 3mo ago
tf is yaml anyway, and llms can output it natively in structured output mode.
362.
by esafak 3mo ago
They could but they don't belong together. A PR should be about one thing. Refactoring one module for another's sake should be done in another PR. If not for reviewing, for revertability, and future debuggability.
363.
by esafak 3mo ago
This is a dated take to me. I think the next stage in unlocking productivity is so called "loop engineering"; figuring out how to effectively not read all the code while ensuring quality. To me that means implementing statistical
364.
by esafak 3mo ago
Because every country wants cheap, expendable, tireless soldiers.
365.
by esafak 3mo ago
https://www.technologyreview.com/2022/12/19/1065306/roomba-i...
366.
by esafak 3mo ago
Why wouldn't you just use Google Maps with a single-$ price filter?
367.
by esafak 3mo ago
From what I gather this marries a self-correcting formal specification based on Signal Temporal Logic (cf. https://en.wikipedia.org/wiki/Temporal_logic ) with a natural language requirement definition (like OpenSpec).
368.
by esafak 3mo ago
What's that, SystemVerilog Assertion?
369.
by esafak 3mo ago
I think that can be algorithmically fixed. Have the AI classify the review by aspects of the app it touches on, then apply the rating to them (say, equally). Then present a multivariate rating, along with a qualitative verbal summary along
370.
by esafak 3mo ago
How about writing with sqlite and querying with duckdb-sqlite, using a replica or WAL, to avoid locks?
371.
by esafak 3mo ago
I'm digging their Sentry charts. That's the kind of discipline we need more of. Stability is a feature.
372.
by esafak 3mo ago
A lot of atomic scientists did regret their work.
373.
by esafak 3mo ago
I like being able to control the visuals with sliders but I have not had much luck replicating their results using GLM. It would be great if each demo showed its prompt.
374.
by esafak 3mo ago
It's not on https://models.dev/ so not in OC.
375.
by esafak 3mo ago
Furthermore, shifting the complexity to the model simplifies engineering. A lot of companies made this trade, replacing their bespoke ML models with foundation models.
376.
by esafak 3mo ago
Because they provide a reasonable common foundation. Otherwise you will forever be arguing with Christian nationalists.
377.
by esafak 3mo ago
https://en.wikipedia.org/wiki/Information_bottleneck_method
378.
by esafak 3mo ago
"You are distilling what I have rightfully pirated."
379.
by esafak 3mo ago
Is it surprising? Every time you log in your IP is recorded, and the ISP can handle the rest.
380.
by esafak 3mo ago
They're selling the robots that use these models. The rest of the world just hasn't got round to embodiment yet.
381.
by esafak 3mo ago
He's reached the singularity. Just kept pointing the agent at its own output, and leveled up...
382.
by esafak 3mo ago
Just make sure you retain the org tree history.
383.
by esafak 3mo ago
People pay a premium for the best.
384.
by esafak 3mo ago
But these models already distill the smarter American ones ;)
385.
by esafak 3mo ago
OpenAI measures token efficiency. Look at the API cost charts in their announcement: https://openai.com/index/gpt-5-6/
386.
by esafak 3mo ago
A poor title for a blog post on someone who just discovered Deepwiki. Here's a shorter and more useful article: https://docs.devin.ai/work-with-devin/deepwiki-mcp
387.
by esafak 3mo ago
If you too are wondering their CI story, it is based on NixOS: https://blog.tangled.org/ci/ https://blog.tangled.org/spindle-microvm/ Curiously the link to the spec is broken: https://t
388.
by esafak 3mo ago
I don't know how you would translate the strength of a robot army to a human one; they haven't fought yet.
389.
by esafak 3mo ago
Analyzing substitutes and complements was a mainstay of data scientists 10-20 years ago. Too mundane to mention now. https://en.wikipedia.org/wiki/Association_rule_learning https://en.wikipedia.org/wiki
390.
by esafak 3mo ago
Have you not heard of things like AlphaFold?

More ›